LEMLEMWORKING PAPER SERIES

Exploring the link between Innovation andGrowth in Chilean irms

Caterina Santi °Pietro Santoleri °

° Institute of Economics, Scuola Superiore Sant'Anna, Pisa, Italy

2016/09 March 2016

ISSN(ONLINE) 2284-0400

Exploring the link between Innovation and Growth in

Chilean firms

Caterina Santi∗

Pietro Santoleri†

Abstract

We employ a balanced panel dataset representative of the entire Chilean productivestructure in order to investigate the relation between the introduction of innovationand subsequent firm growth in terms of sales. Recent contributions examining thereturns to innovation on firm performance have stressed the need of going beyond theanalysis of the ‘average effect for the average firm’. However, previous studies in thecase of Latin American economies have often overlooked the importance of analyzingwhich firms benefit more from the introduction of innovations. Our analysis consistsof a series of parametric and non-parametric exercises which take into account theproperties of the firm growth distribution. In particular, we adopt quantile treatmenteffects (QTE) which allow to estimate the effect of the introduction of innovation bycomparing firms with a similar propensity to innovate for different quantiles of the firmgrowth distribution. On one hand, our results indicate that process innovation showsa positive and significant relation with firm growth for those firms located at the 75thand 90th percentiles. On the other, product innovation shows a negative associationonly for high-growth firms.

Keywords: innovation, firm growth, Chile, quantile regression, quantile treatmenteffectsJEL classification: C14, C21, C22, D22, O31.

1 Introduction

Latin America lags behind the OECD economies in terms of innovation and technologyadoption. Contrarily to East-Asian countries which have narrowed their technological gapsby making their productive structures more complex and knowledge-intensive, Latin Amer-ican economies have made little progress in this area throughout the last decades. Withinthis region, a particular case is represented by Chile. Indeed, despite having the highestGDP per capita in Latin America, this country still features low expenditures on R&D(0.35% of GDP, the 7th economy in Latin America) (OECD/World Bank, 2015). Oneof the reasons for such limited efforts in developing domestic capabilities to generate and

∗Institute of Economics, Scuola Superiore Sant’Anna – c.santi@sssup.it†Institute of Economics, Scuola Superiore Sant’Anna – p.santoleri@sssup.it1Acknowledgments The authors would like to thank Federico Tamagni, Angelo Secchi, Stefano Bian-

chini, Le Li, Nanditha Mathew, Ernest Miguelez for their helpful comments and suggestions. Thanks arealso due to the participants to the DRUID Academy conference at Bordeaux University and to the 3rd Ph.D.workshop in Economics of Innovation, Complexity and Knowledge at Collegio Carlo Alberto in Turin.

1

manage technological change has been the tendency of the business sector not to priori-tise innovation and technological development (OECD/ECLAC, 2013). In fact, the privatesector contributes only for the 32% to the total R&D domestic expenditures, one of thelowest figures worldwide (the average for OECD countries is 68%) (OECD/World Bank,2015). These limited efforts in improving domestic technological capabilities are, in turn,observable in the degree of novelty of the innovations commercialized by Chilean firms. Forinstance, compared with other Latin American countries, Chile has one of the lowest sharesin terms of new-to-the-world product innovations while the bulk refers to new-to-the-firmproducts (Casanova et al., 2011).

Given this context, the aim of this study is to shed more light on the link between theintroduction of technological innovations and firm growth in the Chilean business sector.While previous contributions have focused on the nexus between innovation and produc-tivity (Benavente, 2006; Alvarez et al., 2010, 2012) or employment growth (Benavente andLauterbach, 2008; Alvarez et al., 2011), the link between innovation and growth in terms ofsales at the firm level has not been explicitly investigated in the case of Chile.

Empirical evidence on the expected beneficial effects of innovation is still not conclusiveand calls for further exploration especially in the case of less developed countries. Recentcontributions show that the traditional empirical approach focusing on the ‘average effect forthe average firm’ has mostly yielded inconclusive result, while the use of quantile regressions(QR) has unveiled that innovation is important for a minority of fast-growing firms locatedat the upper quantiles of the growth rate distribution (Moreno and Coad, 2015). Contrarilyto recent studies regarding high-income economies, previous research in the case of LatinAmerican countries did not go beyond the analysis of the returns to innovation for theaverage firm. This represents a significant shortcoming of the existing literature given thehistorical presence of structural heterogeneity (Pinto, 1970; Sunkel, 1978) which is reflectedin huge differentials in terms of firms’ productivity, resources, technological capabilities,access to foreign markets etc. among different firm sizes and between and within sectors(Cimoli et al., 2005; Santoleri and Stumpo, 2011; OECD/ECLAC, 2013; Catela et al., 2015).

Therefore, we employ QR techniques to disentangle how the returns to innovation areassociated with firms located at different quantiles of the firm growth distribution.

However, traditional QR yield estimates through the comparison among potentially verydifferent firms in terms of their propensity to introduce innovation. Thus, we provide ad-ditional evidence exploiting recent econometric developments in quantile treatment effects(QTE) (Firpo, 2007) which allow to estimate the effect of innovation on subsequent salesgrowth by comparing similar firms in terms of their propensity to introduce innovations.

We also assess this relation following Hall and Mairesse (2006) who suggest to examinea longer time span and not just the simultaneous effect of innovation on firm growth.

In order to do that, we exploit a recently published balanced panel data set - EncuestaLongitudinal de Empresas - longitudinally tracking Chilean firms throughout the years 2007,2009 and 2013. Unlike most studies employing Innovation Surveys - where growth andinnovations refer to the same time period due to the cross-sectional structure - our datasetallows to examine the association of innovations with subsequent sales growth at the firmlevel. Also, we must stress that this dataset does not suffer from typical Innovation Survey-bias: that is, the data collection process does not discriminate between innovating or non-innovating firms and therefore we do not have to correct for potential selectivity biases(Mairesse and Mohnen, 2010). Finally, our sample is representative of the whole productivestructure while most of the contributions in this field have examined only manufacturing orrelatively large high/medium-tech firms featured in Innovation Surveys.

2

The results indicate that there is a considerable difference when employing QTE com-pared with traditional QR. We find that process innovation is significantly and positivelyassociated with subsequent sales growth at the 75th and 90th percentiles of the firm growthdistribution. However, we find product innovation to be negatively associated with salesgrowth for those firms at the 90th percentile.

The remainder of the paper is as follows. In the next section we review significantcontributions in the field. In Section 3 we describe the data employed along with somedescriptive statistics regarding the main variables. In Section 4 we describe the econometricprocedure to investigate the relation between innovation and firm growth, while in Section5 and 6 we discuss our empirical results. Finally, Section 7 presents the conclusions.

2 Related Literature

A number of models stemming from different theoretical traditions have considered theimportance of innovation for firm growth (Nelson and Winter, 1982; Jovanovic, 1982; Ericsonand Pakes, 1995; Aghion and Howitt, 2005). However, empirical works examining thisrelation have had difficulties in confirming the theoretical intuitions (Coad and Rao, 2008).Studies such as Geroski et al. (1997), Geroski (2002) and Bottazzi et al. (2001) failed to finda significant effect of innovation on growth while others even yielded a negative one (Freeland Robson, 2004). The reason for this can be related to the very nature of the growth ratedistributions which are characterized by fat-tails (Bottazzi and Secchi, 2006) and by the factthat the average firm doesn’t grow very much entailing the need of examining fast-growingfirms. Indeed, recent contributions employing QR (Koenker and Bassett Jr, 1978) revealthat innovation activities (measured as R&D or patents) do play a significant role for thosefirms located at the top of the firm growth distribution (Coad and Rao, 2008; Coad, 2009;Goedhuys and Sleuwaegen, 2010; Holzl, 2009; Falk, 2012; Colombelli et al., 2013; Bartelsmanet al., 2014; Mazzucato and Parris, 2015).

However, the studies which go beyond the analysis at the mean firm growth have dealtso far with countries at the technological frontier with only few exceptions. Among them,Goedhuys and Sleuwaegen (2010) study manufacturing firms in 11 Sub-Saharan Africancountries and show that product innovation positively affects employment growth for busi-nesses located at the highest percentiles of the firm growth distribution. Damijan et al.(2012), who focuses on Slovenia, find that firms at the bottom of the productivity growthdistribution benefit more from product and process innovation than firms at the top of thedistribution.

Recent contributions have also exploited matching techniques to investigate the effectsof different measures of innovation output on firm performance. These methods allow toestimate the innovation effects by comparing characteristics of similar firms in such a waythat the only difference among them is the successful introduction of innovation. Cozzaet al. (2012), using a sample of medium and high-tech Italian manufacturing firms, find apositive and significant ‘innovation premium’ in terms of profitability and growth for thosefirms which introduced innovative products. Kannebley Jr et al. (2010) investigate the ef-fects of technological innovation on several measures of firm performance for a sample ofmanufacturing firms in Brazil. Their findings indicate that product innovation alone doesnot have a significant impact on performance while process innovation does exert a positiveeffect. However, product innovation appears to have a significant effect when consideringthe degrees of novelty (innovation new-to-the-firm or new-to-the-market) or when jointly

3

undertaken with process innovation. Alvarez et al. (2010) use matching techniques to inves-tigate the effects of introducing a new product for different measures of firm performance inthe Chilean manufacturing industry. Results show that the introduction of a new productdoes not have an effect on productivity or employment growth, while it has a positive oneon current and subsequent sales growth. Moreover, Holzl and Friesenbichler (2010) exploredthe differences in the innovation behaviour of high-growth firms for 16 EU countries. Theresults highlighted that high-growth firms in countries at the technological frontier are moreinnovative and tend to have a higher R&D intensity than average-growth firms, while R&Dand innovation lose relevance as distinguishing features of high-growth and average-growthfirms in countries that are more distant to the technological frontier.

With respect to Latin American firms, studies have mainly drawn upon the so-calledCrepon-Duguet-Mairesse (CDM) model (Crepon et al., 1998) to estimate the contributionof innovation activities on the average firm’s productivity growth. The findings are mixed forLatin American economies: Raffo et al. (2008) find a significant impact of product innovationfor Brazil and Mexico but not for Argentina; Perez et al. (2005), Chudnovsky et al. (2006)do not find significant effect of innovation on firms productivity in Argentinean firms. Onthe other hand, Goedhuys and Veugelers (2012) find that innovation is an important driverfor sales growth in Brazilian firms: in particular, the combination between product andprocess innovations significantly improves firm growth.

In the case of Chile, previous contributions have focused on the links between i) innovation-productivity growth (Benavente, 2006; Alvarez et al., 2010, 2012) and ii) innovation-employm-ent growth (Benavente and Lauterbach, 2008; Alvarez et al., 2010). Regarding the innovation-productivity growth relation for manufacturing firms, Benavente (2006) finds no positive andsignificant effect of innovation on productivity growth while Alvarez et al. (2010) show thatproduct innovation has a significant effect on productivity two years after its introductionbut no simultaneous effect. Alvarez et al. (2012) found that for both sectors technologicalinnovation appears as an important determinants of labor productivity. Concerning inno-vation and employment growth, the evidence suggests that process innovation is generallyfound not to be a determinant of employment growth in Chilean manufacturing plants whileproduct innovation is positively associated with an expansion of employment (Benavente andLauterbach, 2008; Alvarez et al., 2010). In sum, recent empirical evidence found more ho-mogeneous results concerning the impact of innovation on employment growth, while theevidence related to the innovation-productivity growth nexus are less clear-cut for Chileanfirms.

Previous literature has put forward some conjectures regarding the mixed results. Halland Mairesse (2006) suggest that these results may be the consequence of the very dif-fering circumstances under which innovation activities are carried out in these countriesas compared to developed economies and they suggest to examine the effects over longerperiods of time. Bogliacino et al. (2009) and Crespi and Zuniga (2012) maintain that theinnovation process in less developed countries differ from high-income economies: in theformer, stronger R&D capabilities and science and technology infrastructure are essentialto develop the required knowledge and competences to operate at the technological frontier;in the latter, technological change mainly consists of acquiring new machinery and imitateproducts and processes developed elsewhere (Vivarelli, 2014). In other words, innovationactivities carried out in many Latin American countries are incremental with little incidenceon international markets and are mainly based on imitation and technology transfer.

The presence of these specificities concerning the innovation process in less developedcountries may lead to the mismeasurement of innovation output associated with indicators

4

of standard use. According to Cassoni and Ramada-Sarasola (2012) “since new productsdo not constitute the core of innovation in non-developed economies, standard innovationoutput indicators - the share of innovative sales; the accumulated number of patents - wouldnot act as good proxy variables whenever most firms are process-innovative” (p.143).

With respect to Chile, Alvarez et al. (2010) claims that the empirical evidence “might beconsistent with a very slow process of learning by doing in the mastering of new productionprocesses on the part of Chilean firms. These slow and, most of the time uncertain, gains inproductivity could help to explain the low levels of investment in R&D activities by Chileanfirms”.

3 Data

In order to conduct the study, we exploit three waves of a rich dataset released by theMinisterio de Economia (2009, 2012, 2015): the Encuesta Longitudinal de Empresas1 (ELE).In each wave a part of the sample is fixed while the other part is renewed every time.This permits analyzing variations over time for those firms observed in different waves. Inorder to implement the empirical strategy outlined in Section 4, we merged observationsregarding firms appearing in each wave. By doing this, we obtain a balanced panel of 1,839formal businesses stratified by economic sectors (1 digit ISIC rev. 3 codes) and firm size(considering the level of sales) covering the years 2007, 2009 and 2013 (t = 1, 2, 3)2. We haveto highlight two features of this survey which are important for the outcome of our analysis:it represents formal businesses and includes micro enterprises starting from 800 UF worthannual sales3. As previously mentioned, the sample allows us to get a complete pictureof the entire productive structure than other available surveys (see for instance ENIA formanufacturing firms and the Chilean Innovation Survey focused mainly on high/mediumtech firms). Also, we must stress that this dataset does not suffer from typical InnovationSurvey-bias: that is, the data collection process does not discriminate between innovating ornon-innovating firms and therefore we do not have to correct for potential selectivity biases(Mairesse and Mohnen, 2010).

We cleaned the data in the following way: first, we selected only firms with at least oneemployee; second, we eliminated firms with missing sales; third, we restricted observationsto those firms with an average annual sales growth lower than 500% in order to controlfor the presence of growth processes derived from mergers and acquisitions and anomaliesin the data4. In the end, we are then left with a sample of 1,668 firms observed for twotime periods (see Table 3 for the structure of the dataset). Table 8 in the Appendix showsthe distribution of the unweighted sample across industries and firm sizes which is mainlycomposed by small (27 per cent) and large businesses (45 per cent) in the Retail (20 percent) and Manufacturing (15 per cent) sectors.

1This survey has been little exploited so far with some exceptions being Alvarez and Crespi (2015), Cirillo(2014) and Santoleri (2015).

2Since the third wave of the ELE provides firm revenues for 2012 in the 3rd wave, we employ 2012 insteadof 2013 in the analysis in order to have more homogeneous time intervals.

3It corresponds approximately to 31,000 USD (1 UF = to 39 USD). Firm sizes are stratified according tothe following segmentation: micro (800-2,400 UF), small (2,400-25,000 UF), medium (25,000-100,000 UF)and large (100,000 or more).

4This cleaning procedure concerning the growth rates led to the exclusion of 57 firms. Since the settingof the 500% threshold could be considered somewhat discretionary, we conducted the analysis also using alower and a higher threshold of, respectively, 400% and 600% and results do not change.

5

Table 1: The structure of the balanced panel

Year Time Growtht Innot−1

2007 T1 - Inno12009 T2 G1 Inno22012 T3 G2 -

3.1 Firm Growth

Following previous literature, we define the annual average firms’ growth rate as:

Gri,t =ln(Si,t)− ln(Si,t−d)

d, (1)

where Si,t is the firm sales at time t and Si,t−d is its value d years before5. Sales are deflatedusing the Chilean GDP deflator (base year 2008) drawn by the Central Bank database.

Figure 1 shows the distribution of firm growth rates. In line with other empirical con-tributions (Bottazzi and Secchi, 2006; Coad, 2009), we can observe that sales growth arenot normally distributed. Indeed, they exhibit excess values close to the mean and fat-tailswhich prevent using Gaussian assumptions. Hence, the properties of the data motivate usto adopt QR and QTE approach in the empirical analysis.

Figure 1: Empirical firm growth distribution

0.0001

0.001

0.01

0.1

1

10

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Empirical densityAEP fit

NORMAL fit

Source: Authors’ elaboration based on ELE. Growth computed with eq.1. Note the y-axis is on log-scale. The

parameter br is 0.75 and bl is 0.66 confirming the non-normal distribution (see Bottazzi and Secchi (2011) for

more details on the AEP distribution).

5As Audretsch et al. (2014) pointed out, the indicators used to measure growth are not neutral withrespect to empirical results. Different measures describe different patterns of growth (Delmar et al., 2003).For this reason, we introduce an alternative growth rate measure to conduct a robustness check, the resultsare shown in Section 8.1

6

3.2 Innovation variables

The ELE provides data on a wide range of issues. With respect to technological innova-tion, drawing on the OSLO Manual (OECD, 2005), the ELE defines product and processinnovation as follows:

• A product innovation (InnoProd) is the introduction of a good or service that isnew or significantly improved with respect to its characteristics or intended uses.This includes significant improvements in technical specifications, components andmaterials, incorporated software, user friendliness or other functional characteristics.

• A process innovation (InnoProc) is the implementation of a new or significantly im-proved production or delivery method. This includes significant changes in techniques,equipment and/or software.

In order to investigate the relation between these activities and subsequent firm growth,we employ dichotomous variables indicating whether or not a firm implemented an innova-tion during one of the years covered by the survey 6.

As shown in Table 8 in the Appendix, firms successfully innovating in product and processare, respectively, 27 and 24 per cent. Those figures are in line with the results of the ChileanInnovation Survey: the innovation rate, that is, the percentage of firms which carried outsome kind of innovation, is 23.7% for the years 2011-2012 (Ministerio de Economia, 2014).

Unfortunately, the dataset does not feature any additional information regarding thequality of the innovation, e.g. if it is new-to-the-firm or new-to-the-market, or the outcome ofthe introduction of an innovation on business performance, e.g. share of innovative productsin firm’s turnover. Although the ELE does not provide more detailed data regarding thequality of the innovation outcome in Chilean firms, we should bear in mind that in lessdeveloped countries, the very nature of innovation differs from countries at the technologicalfrontier. While in the latter the introduction of a new product often means placing it onthe world market, in less developed countries the degree of novelty generally extends onlyto domestic markets since firms display a very limited capacity in terms of in-house R&D(Hall and Mairesse, 2006; Bogliacino et al., 2009). Indeed, if we compare data from theChilean Innovation Survey (Ministerio de Economia, 2014) to the ones of the EuropeanCommunity Innovation Survey (Eurostat, 2013) we can observe a significant difference inthe degree of innovation novelty: in Chile only 34 per cent of the product innovations arenew to the market, whereas in countries such as France, Italy, Norway and the Netherlandsthose reach, respectively, 66, 61, 79 and 67 per cent. Moreover, as previously mentioned,the degree of novelty of the innovations commercialized by Chilean firms appear to berelatively low even when compared with other Latin American economies (Casanova et al.,2011). These considerations are important since it is likely that our explanatory variable iscapturing product innovation characterized by a relatively low degree of novelty.

3.3 Control Variables

Several other factors are likely to influence the propensity of a firm to grow. Thus, inorder to examine how and whether innovation influences growth, one needs to control for

6Wemust stress that, while the 2nd and 3rd waves feature detailed information concerning R&D activities,the 1st one collects data on R&D which are not strictly comparable with those present in other waves and,therefore, we restrain from investigating the association between conducting R&D and subsequent firmgrowth.

7

alternative determinants of firm performance. According to Gibrat’s law (Gibrat, 1931),firm growth rates are independent of firm size. However, empirical evidence shows thatsmaller and younger firms have higher expected growth rates than older and larger firms(Mansfield, 1962; Coad, 2009). Therefore firm size is included as an explanatory variableexpressed as the logarithm of the number of employees (Employment). We control for firmage (Age) since previous contribution have highlighted its relevance for firm performance(Coad, 2009; Coad et al., 2013; Haltiwanger et al., 2013). According to Pavitt (1984),the propensity to innovate varies among industries because technological opportunities areuneven across economic activities. Following this approach, 12 sectoral dummies (ISICrev. 3) are employed in our regression analysis to control for potential sectoral systematicdifferences. Moreover, the export status of a firm could act as a driver of higher growthand therefore we include a variable identifying the intensity of exports by Chilean firmsmeasured as the percentage of exports over total sales (Expint). The degree of foreignownership is another factor potentially affecting growth as it offers businesses wider accessto skilled staff and potentially enables businesses to respond more quickly to opportunitiesoverseas. However, empirical results have shown that the relation is ambiguous and dependson the type of FDI and the country of origin of the firm (Chudnovsky et al., 2006). Wecapture foreign ownership with a dummy variable measuring whether part of the firm’scapital is foreign owned (Foreign). Previous studies have emphasized the importance ofmarket concentration within sectors as an important determinant of firm growth (Geroski,1995). Indeed, there might be considerable barriers to entry and growth in industries withhigh market concentration, where large incumbents might engage in strategic behaviour toprevent growth of smaller firms (Geroski, 1995; Nickell, 1996; Lever and Nieuwenhuijsen,1999). For this reason, we construct the Herfindahl Index (Herfin) in order to controlfor concentration at the sectoral level calculated by the sum of the squared firm marketshare within each industry. We also include the percentage of employees with tertiaryeducation (HC ) as a proxy of human capital since this is expected to positively stimulategrowth (Goedhuys and Sleuwaegen, 2016). Finally, we insert a dummy variable measuringwhether the firm is part of a business group (Group) and a dummy indicating whether afirm has received public support through productive development policies implemented bythe Chilean Government (Public).

Table 2 presents the average sales growth conditional upon past innovation activities andthe average of the control variables conditional upon contemporaneous innovation activitiesfor the pooled sample. On average, firms that innovate have higher levels of sales in thesame period and higher rates of sales growth in the following period. The only exceptionis represented by those firms conducting product innovation: although they have a higherlevel of sales, firms that introduced this type of innovation at t− 1 experience lower growthrates at time t on average. Moreover, innovating firms display a considerably larger size(in terms of employment) along with a larger share of employees with tertiary education,higher export intensity (measured as exports over total sales) and they are generally olderthan non-innovating firms. Additionally, innovators are more often part of a business groupand owned by foreign capital, they more frequently benefit from public subsidies and, asobserved by Geroski and Machin (1992), generally operate in sectors with slightly higherconcentration.

8

Table 2: Mean values for innovating and non-innovating firms

Product Innovatorst−1 Process Innovatorst−1 All firmsNo Yes No Yes

Salest−1 25,520 57,694 20,201 78,828 34,208Growtht 0.023 0.021 0.018 0.037 0.023HCt−1 16.57 20.48 17.00 19.68 17.64Foreignt−1 7.67 12.58 7.64 13.39 9.00Expintt−1 4.24 6.50 4.22 6.96 4.88Groupt−1 23.82 37.96 23.05 42.42 27.69Employmentt−1 288 652 268 768 388Aget−1 16.42 18.98 16.28 19.78 17.12Publict−1 15.31 25.71 14.15 30.91 18.16Herfindahlt−1 0.09 0.10 0.09 0.10 0.09

N 2422 914 2537 799 3336

Source: Authors’ elaboration based on ELE. Note that Sales is in millions of Chilean Pesos while Growth isexpressed in logarithmic terms. Expint, Foreign, Group and Public are expressed in percentages.

4 Methodology

We start investigating the relationship between sales growth and innovation by comparingthe growth rate distribution among innovators and non-innovators along each innovationvariable. We perform a two-sample Fligner-Policello (FP) robust rank order test (Flignerand Policello, 1981) to check if the sales growth rates of innovators and non-innovatorsare sampled from the same population. This test assumes that the groups of innovatorsand non-innovators are independent samples from continuous distributions symmetric withrespect to the population medians (θI , θN ), the null hypothesis is H0 : θI = θN .

Successively, in order to further delve into the examination of the links between differenttypes of innovation and firm growth, our empirical strategy entails the use of parametricand non-parametric methods illustrated in the three following sections.

4.1 Quantile Regression Analysis

As a first step we estimate the following augmented Gibrat’s Law equation using OrdinaryLeast Squares (OLS) and Fixed-Effects (FE) with robust errors using Huber-White sandwichestimator and QR with boostrapped standard errors7:

Gri,t = α+ β1Innoi,t−1 + ~λXi,t−1 + ui,t (2)

where the dependent variable Gri,t is sales growth of firm i in year t, measured as in (1);Innoi,t−1 identifies the following variables: InnoProdi,t−1 which represents lagged productinnovation and InnoProci,t−1 lagged process innovation. Additionally, Xi,t−1 represents aset of lagged control variables including, among others, firm size and firm age.

It is important to stress the rationale behind the econometric procedures employed in ourexercise. Given that sales growth distribution are fat-tailed and skewed (see Figure 1), OLSand FE would produce biased estimates. Instead, the QR approach (Koenker and Bassett Jr,1978) provides a more robust and efficient estimation method in comparison to OLS and FE

7In order to check the robustness of the QR estimates, we employed a recent method proposed by Parenteand Santos Silva (2016) which allows to obtain standard errors asymptotically valid under heteroskedasticityand misspecification. The estimates are similar whether we estimate the asymptotic covariance matrix orwe estimate the covariance matrix by bootstrap. The results are available upon request.

9

in the presence of outliers and when the error term is not normally distributed (Buchinsky,1998). Specifically, QR allow to obtain a more detailed picture of the heterogeneous returnsto innovation. As a result, our preferred specification will be the QR, while OLS and FEregressions are anyhow presented for completeness.

An important assumption underlying our empirical strategy is that there is a time lagin the growth response due to the introduction of innovations. This conjecture stems fromempirical evidence also in the case of Chile (Alvarez et al., 2010). Furthermore, the prob-lem of reverse causality is partially mitigated with the introduction of lagged innovationswhile growth refers to the successive period. Hence, the effects of successful innovationson firm performance are examined after their introduction over two different time horizons,respectively, two and three years.

Finally, it is important to stress that the estimates presented in the following sectionscorrespond to associations between firms’ growth and the different types of innovation ratherthan estimates of causal effects.

4.2 PSM and NNM Analysis

In order to have a richer understanding of the relations examined in our study, we com-plement the previous analysis by estimating the average treatment effect (ATE) and theaverage treatment effect on the treated (ATET). In order to do this, we propensity scorematching (PSM) and nearest neighbours matching (NNM) estimators following recent con-tributions analysing innovation and firm performance (Kannebley Jr et al., 2010; Alvarezet al., 2010; Coad et al., 2015). These methods allow to estimate the effect of innovation onsales growth by comparing firms with a similar propensity to introduce innovations.

The PSM consists of two stages: in the first one a logit model for each of the innovationvariables is estimated in order to build our counterfactual sample. Conditional on satisfyingthe balancing property of the propensity score, the fitted values obtained from estimating thelogit estimation are used to pair up innovators with non-innovators and those matched pairsare subsequently used to estimate the average treatment effect of innovation on firm growth.The PSM requires that all variables relevant to the probability of receiving treatment maybe observed and included in X. Moreover, in order to find adequate matches, it is necessaryto ensure a sufficient overlap in the characteristics of treated and untreated units. Whilethe last assumption can be easily tested, the first one is difficult to comply with.

The NNM method imputes the missing potential outcome for each subject by using anaverage of the outcomes of similar subjects that receive the other treatment level. Similaritybetween subjects is based on a weighted function of the covariates for each observation. NNMtechnique is non-parametric, hence it requires a large number of observations to get to thetrue treatment effect vis-a-vis an estimator that imposes a functional form8.

For both techniques, the matching procedure between innovators and non-innovatorsis based on the entire set of control variables discussed in Section 3.3: sectoral dummies,firm size (in terms of employment), firm age, foreign ownership, whether they are part of abusiness group, share of employees with tertiary education, export intensity, whether theyreceived public subsidies, year dummies and the Herfindahl index.

8Abadie and Imbens (2006, 2011) show that nearest-neighbor matching estimators are not consistent whenmatching on two or more continuous covariates. Following their contribution we introduce bias-correctionto fix this problem Abadie and Imbens (2012).

10

4.3 Unconditional Quantile Treatment Effects Analysis

The matching techniques presented so far estimate the average effects of the introductionof innovation. Since our goal is to take into account the heterogeneity of innovation returnsand the structural differentials in firm capabilities which characterize Chilean productivestructure, we also estimate unconditional QTE using matching techniques (Firpo, 2007).

This method describes the difference in the quantiles of the outcome variable (firmgrowth) for innovators and non innovators without reference to the control variables. Indeed,covariates are used only to estimate the propensity score of the probability of introducingan innovation, thus allowing to compare similar firms. Specifically, the propensity-scoreweighting estimator is obtained non-parametrically by a local logit estimator from Innoi,given X and a constant. Then, the estimator of the QTE parameter is ∆θ = qθ1 − qθ0 , wherefor I ∈ {0, 1},

(α, qθI ) = argminα, q

{

1

N

N∑

i=1

ωi,Iρθ(Gri,t − α− qInnoi,t−1)

}

, (3)

where

ωi,1 =Innoi,t−1

P r[Innoi,t−1 = 1|Xi,t−1]and ωi,0 =

1− Innoi,t−1

1− P r[Innoi,t−1 = 1|Xi,t−1], (4)

are the weights, and P r[Innoi,t−1 = 1|Xi,t−1] is the nonparametric estimator of the propen-sity score.

As illustrated by Frolich and Melly (2010), the definition of unconditional QTE doesnot change when we change the set of covariates X. Contrarily, standard QR allows therelationship between the explanatory variable and the dependent variable to vary based ona nonseparable disturbance term labelled “unobserved proneness” for the outcome variable.As covariates are added, some of this unobserved proneness becomes observed and, con-sequently, the interpretation of the estimates changes. In other words, take the examplerelating firm growth to firm age. In this case, the unconditional 90th percentile refers tohigh-growth firms, whereas the 90th percentile conditional on age refers to high-growth firmswithin each age class, who however may not necessarily be high-growth overall. Therefore,the interpretation of the 90th quantile is different if one considers conditional and uncondi-tional quantiles.

A shortcoming of the unconditional QTE estimator by Firpo (2007) is that it relieson the assumption of exogeneity of the treatment variable. Hence, although QTE shouldprovide more accurate estimates compared with traditional QR, we cannot claim that theresults correspond to causal effects.

5 Results

As a first step we compare the distributions of growth rates for innovating and non-innovatingfirms by means of the Fligner-Policello tests (results are displayed in Table 3). We can seethat for those firms introducing process innovation the underlying sample distribution is notthe same as the one for the firms which do not innovate. Moreover, given that in case ofrejection of the null hypothesis, the sign of the FP statistics (U) tells which of the two groupis dominant: a negative (positive) sign means that innovator (non-innovator) firms have a

11

higher probability to get higher sales growth rates. We can state that the group of innova-tors stochastically dominates non-innovator firms. Differently, the test does not reject thenull in the case of product innovators implying that there is no difference in the distributionof growth rates for firms that introduced or not introduced that kind of innovation. Theseresults confirm the hint given by descriptive statistics which showed that firms carrying outproduct innovation experienced almost the same growth than the rest.

Table 3: Two-sample Fligner-Policello robust rank order test

Variable Obs Mean Index of Variability U 2-Tailed p-value

InnoProdt−1 No 2422 4.5e+02 1.9e+08 -0.392 0.69495InnoProdt−1 Yes 914 1.2e+03 4.0e+08

InnoProct−1 No 2537 3.8e+02 1.4e+08 -2.502 0.01235InnoProct−1 Yes 799 1.3e+03 4.3e+08

Source: Authors’ elaboration based on ELE. Growth is defined as in (1).

After testing for whether innovation can be considered a source of growth differentials, weanalyze the effects of innovative activities on firm growth using QR and matching techniques.

5.1 Quantile Regression Estimates

In Table 4 we report OLS, FE and QR estimates of eq. (2). First of all, we can observethat according to OLS estimates product innovation is not significantly different from zero.However, process innovation is found to be positively and significantly (5 per cent confidencelevel) related with firm growth. In the FE specification, the point estimate of productinnovation is negative and significant at 5 per cent level, while process innovation coefficientis positive but not significant.

However, QR yield quite interesting results. Indeed, we can observe that, while prod-uct innovation is not significant, process innovation yields positive and significant results.Although OLS estimates were able to capture the importance of process innovation for theaverage firm, QR reveal the existence of a U-shaped relation between this type of innovationand subsequent firm growth: the importance of process innovation increases when consider-ing both the lower and upper tails of the growth distribution. This appears to be consistentwith the incremental nature of the innovation activity in less developed countries which ismore process-oriented than product-oriented as pointed out by several studies (Cassoni andRamada-Sarasola, 2012; Crespi and Zuniga, 2012).

Concerning the control variables, the QR results for firm size and age are broadly inline with previous contributions which show that smaller and younger firms have higherexpected growth rates than older and larger firms (Mansfield, 1962; Coad, 2009). Indeed,the coefficient on lagged size (in terms of employment) is statistically different from zeroshowing a decreasing association with firm growth. Second, age negatively affects firmgrowth for the highest quantiles, with a strong significance level, confirming the intuitionthat younger firms are typically growing more rapidly than older and more mature firms.Third, we observe that export status has a negative and significant coefficient for the lowertail of the distribution. This maybe unexpected, since the literature on micro-empiricsof exports suggest that exporters typically reach superior performance than non-exporters.Recall, however, that a similar result is found for Spanish firms in Holzl (2009) and Bianchiniet al. (2016). As in Goedhuys and Veugelers (2012), we do not find tertiary education havinga positive and significant impact on firm growth.

12

Table 4: OLS, FE and QR - Innovation and Sales Growth(1) (2) (3) (4) (5) (6) (7)OLS FE q10 q25 q50 q75 q90

InnoProdt−1 0.013 -0.038** -0.037 0.001 -0.005 -0.002 -0.022(0.012) (0.018) (0.023) (0.011) (0.007) (0.009) (0.014)

InnoProct−1 0.025** 0.017 0.050** 0.017 0.015* 0.026*** 0.042***(0.012) (0.018) (0.020) (0.012) (0.008) (0.010) (0.016)

Employmentt−1 0.002 -0.065*** 0.020*** 0.012*** 0.005*** -0.003 -0.011***(0.003) (0.013) (0.005) (0.003) (0.002) (0.002) (0.004)

Aget−1 -0.025*** -0.020 0.018 0.010 -0.010** -0.037*** -0.065***(0.008) (0.023) (0.016) (0.008) (0.005) (0.006) (0.011)

Foreignt−1 -0.003 0.011 0.004 -0.010 0.007 0.030** 0.053**(0.019) (0.039) (0.036) (0.021) (0.013) (0.015) (0.025)

Expintt−1 -0.001* -0.001 -0.001*** -0.001** -0.000** -0.000 0.001(0.000) (0.001) (0.001) (0.000) (0.000) (0.000) (0.001)

Groupt−1 0.009 -0.049 -0.029 -0.023* -0.009 0.009 0.024(0.013) (0.032) (0.025) (0.012) (0.008) (0.010) (0.016)

HCt−1 -0.000 -0.001 -0.000 0.000 0.000 0.000 -0.000(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Publict−1 0.018 0.015 0.036* 0.017 -0.003 0.002 -0.003(0.011) (0.020) (0.020) (0.011) (0.007) (0.010) (0.017)

Herfint−1 -0.008 -0.147 -0.122 -0.112 0.037 0.018 -0.647(0.285) (0.302) (0.620) (0.342) (0.178) (0.218) (0.464)

Constant 0.036 0.360*** -0.451*** -0.235*** -0.013 0.199*** 0.602***(0.052) (0.121) (0.104) (0.050) (0.033) (0.044) (0.083)

Observations 3,336 3,336 3,336 3,336 3,336 3,336 3,336R2 0.009 0.042 0.044 0.022 0.006 0.021 0.052

Notes: Robust standard errors and boostrapped estimates (1500 iterations). ***, ** and * indicatesignificance on a 1%, 5% and 10% level, respectively. Industry and time dummies are included but not

reported.

In sum, what we can conclude from the QR estimates is that product innovation does notsignificantly affect firms performance while process innovation shows a U-shaped relationshipwith firm growth particularly affecting firms at the lower and upper parts of the distribution.

5.2 PSM and NNM Estimates

In order to provide further evidence, we employ matching techniques: firm propensity toinnovate is used to match innovating firms with otherwise similar non-innovating firms toevaluate the importance of innovation for firm growth.

In Table 5, we report the ATET and the ATE obtained by PSM and NNM9. The ATETconfirms previous results concerning the importance of process innovation for subsequentfirm growth. Moreover, product innovation has a negative and significant (10 per centconfidence level) impact on subsequent sales growth when we use NNM. The ATE shows

9As previously mentioned, treatment-effects estimators reweight the observational data in hopes of achiev-ing experimental-like balanced data results. If the reweighting is successful, then the weighted distributionof each covariate should be the same across treatment groups. In such cases, we say that the treatmentmodel ”balanced” the covariates. In order to check whether our control groups were balanced we run twodiagnostic tests for both PSM and NNM: i) for each covariate we examined the model-adjusted differencein means in the treatment groups and the ratio of variances; ii) we graphed the model-adjusted estimatedprobability density functions and the box plots of each covariate over treatment levels for the raw data andfor the matched sample. When we consider the NNM, the differences in weighted means are negligible (allclose to zero), and variance ratios are all near one indicating that matched data appear to be balanced.Likewise, for the NNM, graphical representation of probability density functions and box plots also indicatecovariate balance. On the contrary, the PSM performs worse in terms of balancing when checking both themodel-adjusted differences and the box plots for each co-variate. Therefore, we consider the NNM to bemore accurate estimates since they show a better covariate balance.

13

similar results compared to the ATET estimates. Although, product innovation is neversignificant.

Table 5: Average Matching EstimatesAverage Treatment Effects on the Treated - PSM

Coefficient AI SE z P > |z|

Product Innovators -0.00027 0.0159 -0.02 0.986Process Innovators 0.02471* 0.0139 1.77 0.077

Average Treatment Effects on the Treated - NNMCoefficient AI SE z P > |z|

Product Innovators -0.02491* 0.0151 -1.65 0.099Process Innovators 0.03692** 0.0161 2.29 0.022

Average Treatment Effects - PSMCoefficient AI SE z P > |z|

Product Innovators -0.007104 0.0152 -0.47 0.641Process Innovators 0.031042* 0.0184 1.68 0.092

Average Treatment Effects - NNMCoefficient AI SE z P > |z|

Product Innovators -0.02083 0.0147 -1.41 0.157Process Innovators 0.05295*** 0.0157 3.36 0.001

Notes: ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Estimates for the NNMare obtained using the Abadie and Imbens (2006, 2011) bias-corrected estimator for robust standard errors (AISE). Treatment variables are innovation dummies. The selection of the control group is based on the entireset of control variables discussed in Section 3.3: sectoral dummies, firm size (in terms of employment), firmage, foreign ownership, whether they are part of a business group, share of employees with tertiary education,export intensity, whether they received public subsidies, the Herfindahl index and year dummies. N = 3341, 1minimum required match for NNM estimates and 2 minimum required matches for PSM estimates.

5.3 Unconditional QTE Estimates

In Table 6, we complement the matching analysis on the average effects with QTE in orderto examine the heterogeneity of innovation returns to firm growth. This method allows for amore accurate assessment of the relation under study since it compares similar firms in termsof their probability of introducing innovation and estimate these effects taking into accountthe unconditional distribution of the outcome variable (i.e. sales growth). The resultslargely confirm the importance of process innovation, while product innovation is negativelyand significantly (10 per cent confidence level) associated with firm growth for the highestquantile of the distribution. However, in contrast with traditional QR estimates, we donot find the same U-shaped relation between process innovation and firm growth. Indeed,process innovation seems to significantly and positively affect only fast-growing firms (75thand 90th percentiles). Moreover, the patterns of significance and point estimates indicatethat process innovation appear to have a stronger relevance at the 75th percentile than atthe 90th percentile.

14

Table 6: Unconditional Quantile Treatment Effect Estimatesq10 q25 q50 q75 q90

Product Innovators -0.0067 0.0074 -0.0042 -0.0132 -0.0356*(0.0316) (0.0138) (0.0084) (0.0111) (0.0178)

Process Innovators 0.0138 0.0215 0.01826 0.0492** 0.0433*(0.0269) (0.0191) (0.0120) (0.0167) (0.0241)

Notes: Standard errors in parenthesis. ***, ** and * indicate significance on a 1%, 5% and 10% level, respec-tively. The first-step regression to calculate the propensity score is obtained using ivqte while unconditionalQTE are obtained with the STATA command ivqte both developed by Frolich and Melly (2010). The selectionof the control group is based on the entire set of control variables discussed in Section 3.3: sectoral dummies,firm size (in terms of employment), firm age, foreign ownership, whether they are part of a business group, shareof employees with tertiary education, export intensity, whether they received public subsidies, the Herfindahlindex and year dummies. N = 3341.

6 Discussion

The results presented in the previous section confirm the importance of adopting estimationmethods that take into account the heterogeneity of firms’ performance. Indeed, firms’ salesgrowth is typically characterized by heavier tails than the Gaussian distribution. Moreover,accounting for the heterogeneity of the returns to innovation is especially relevant whenconsidering Latin American countries since the productive structures are characterized bysignificant structural differentials among firm sizes and within and between sectors (Catelaet al., 2015). Thus, QR techniques provide a more suited alternative to OLS and FE giventhe properties of the data. According to these estimates, we see that process innovationshows a U-shaped relation with firm growth, affecting especially the lowest and the highestpercentiles.

However, we argue that it is crucial to obtain estimates of the innovation effects bycomparing similar firms in such a way that the only difference among them will be thesuccessful introduction of innovation. Hence, we choose the unconditional QTE under ex-ogeneity (Firpo, 2007; Frolich and Melly, 2010) as our preferred method. Indeed, the esti-mates deriving from QR may come from the comparison of very different firms. Contrarily,the unconditional QTE approach allows to estimate the effects of introducing an innovationbased on the comparison between similar firms using matching techniques.

The results based on the QTE approach do show important differences if compared withstandard QR. On one hand, the U-shaped relation of process innovation and firm growthdisappears: process innovation positively affects only fast-growing firms. On the other,product innovation turns up to be significantly and negatively associated with firm growthfor those business located at the 90th percentile.

The lack of positive association between product innovation and firm growth, althoughat odds with theoretical expectations, is not an unprecedented finding in literature. Indeed,studies such as Holzl (2009), Kannebley Jr et al. (2010), Holzl and Friesenbichler (2010)do not find evidence of product innovators necessarily performing better than non-productinnovators or they even find a negative relation (Freel and Robson, 2004). In particular,our contribution is in line with those studies which do not show a significant association forproduct innovation in emerging countries (Kannebley Jr et al., 2010), or those taking intoconsideration countries which are distant to the technological frontier (Holzl, 2009; Holzland Friesenbichler, 2010; Damijan et al., 2012). Indeed, innovation expenditures in lessdeveloped countries generally focus on the acquisition of foreign machinery and equipmentthat embody newer vintage of technology (which relates to process innovation) rather than

15

on strong R&D in-house capabilities (which relate more with product innovation) (Vivarelli,2014).

Nonetheless, we can point at three additional explanations regarding the results onproduct innovation. First, as previously highlighted, this might be related with the waywe measure this innovation activity: indeed, the ELE does not provide any additionalinformation regarding the ‘quality’ of product innovation or its share on firm’s turnover.In fact, previous studies have highlighted the importance of taking into account the degreeof novelty of product innovation when assessing its effect on sales growth (Duguet, 2006;Kannebley Jr et al., 2010; Cucculelli and Ermini, 2012; Bianchini et al., 2016). Therefore,our imperfect measure of product innovation might fail to isolate those ‘radical’ innovationswhich might be more beneficial in terms of firm performance since they do not refer topreviously existing products and because no competitor produces it (Duguet, 2006).

Second, we should bear in mind that in less developed countries, the very nature ofinnovation differs from developed economies. While in the latter the introduction of a newproduct usually means placing it on the international market, in less developed countriesthe degree of novelty generally extends only to domestic markets since the underlying inno-vations are the result of technology transfers and adoption from abroad (Hall and Mairesse,2006; Bogliacino et al., 2009).

These considerations are especially relevant in the case of Chile. Indeed, the limitedefforts in improving domestic technological capabilities by the private sector affect the degreeof novelty of the innovations commercialized by Chilean firms: when compared with otherLatin American countries, Chile has one of the lowest shares in terms of new-to-the-worldproduct innovations while the bulk refers to new-to-the-firm products (Casanova et al.,2011).

Third, we cannot rule out the possibility that our results could be partly driven bythe 2008-2009 global crisis although Latin American countries and Chile were less affectedthan others and GDP recovered already in 201010. Indeed, innovative efforts tend to beprocyclical and decrease significantly during recessions (Paunov, 2012) motivated by lowdemand expectations. Therefore, a possible explanation for our results is that firms engagedless intensively in product innovation projects because of weak demand expectations fortheir products and dedicated more effort to achieve efficiency improvements through processinnovation.

7 Conclusions

In summary, this article sought to investigate the relationship between innovation (in bothproducts and processes) and subsequent firm performance in terms of sales growth employinga sample representative of the Chilean productive structure. To this end, the most relevantfindings highlight a positive relationship between process innovation and growth for thosefirms located at the upper-tail of the firm growth distribution. Moreover, we find a negativerelationship between product innovation and growth in sales only for high-growth firmswhile this innovation activity is not significant for the rest of the distribution. Overall, theresults confirm that innovation as a highly uncertain activity, with the returns to innovationbeing remarkably heterogeneous with some firms benefiting more than others (Coad, 2009).

The study is bound by several limitations. First, innovation processes are captured byimperfect measures. For instance, it would be interesting to investigate the relation between

10See Paunov (2012) on the impact of the crisis on innovation activities in Latin American firms.

16

firm growth and the novelty of the innovations introduced by Chilean firms as this couldshed more light on the object of analysis. Second, we cannot rule out potential endogeneitybias in our estimates although the lag structure should mitigate this problem.

Future research should investigate whether the persistence in innovating behaviour andthe combination of different types of innovation have a larger impact on subsequent firms’performances. Indeed, Damanpour et al. (2009) suggest that combinations of different typesof innovations may have a larger impact on firms performances. Moreover, one could testthe hypothesis according to which firms benefit from their innovations through increasedprice-cost margins (Cohen and Klepper, 1996) or profit margins (Geroski and Machin, 1992)rather than higher sales growth. Finally, by employing panel data featuring a longer timespan such as the Chilean manufacturing survey (ENIA), it would be possible to investigatethe heterogeneity of the impacts of the successful introduction of product innovation on theconditional distribution of firm growth rates using a Fixed-Effects Panel Quantile Regressionapproach (Canay, 2011).

References

Abadie, A. and G. Imbens (2012). Matching on the estimated propensity score. harvard university andnational bureau of economic research.

Aghion, P. and P. Howitt (2005). Growth with quality-improving innovations: an integrated framework.Handbook of economic growth 1, 67–110.

Alvarez, R., J. M. Benavente, R. Campusano, and C. Cuevas (2011). Employment generation, firm size,and innovation in chile. Technical report, Inter-American Development Bank.

Alvarez, R., C. Bravo-Ortega, and L. Navarro (2010). Innovation, r&d investment and productivity in chile.

Alvarez, R. and G. A. Crespi (2015). Heterogeneous effects of financial constraints on innovation: Evidencefrom chile. Science and Public Policy, scu091.

Alvarez, R., A. Zahler, and C. Bravo Ortega (2012). Innovation and productivity in services: Evidence fromchile. Technical report, Inter-American Development Bank.

Audretsch, D. B., A. Coad, and A. Segarra (2014). Firm growth and innovation. Small Business Eco-

nomics 43 (4), 743–749.

Bartelsman, E., S. Dobbelaere, and B. Peters (2014). Allocation of human capital and innovation at thefrontier: Firm-level evidence on germany and the netherlands. Industrial and Corporate Change, dtu038.

Benavente, J. M. (2006). The role of research and innovation in promoting productivity in chile. Economics

of Innovation and New Technology 15 (4-5), 301–315.

Benavente, J. M. and R. Lauterbach (2008). Technological innovation and employment: complements orsubstitutes? The European Journal of Development Research 20 (2), 318–329.

Bianchini, S., G. Pellegrino, and F. Tamagni (2016). Innovation strategies and firm growth: New longitudinalevidence from spanish firms. LEM Working Paper, Sant’Anna School of Advanced Studies, Pisa, Italy.

Birch, D. G. (1987). Job creation in america: How our smallest companies put the most people to work. Uni-

versity of Illinois at Urbana-Champaign’s Academy for Entrepreneurial Leadership Historical Research

Reference in Entrepreneurship.

Bogliacino, F., G. Perani, M. Pianta, and S. Supino (2009). Innovation in developing countries. the evidencefrom innovation surveys. In FIRB conference Research and Entrepreneurship in the Knowledge-based

Economy, Milano: Universita L. Bocconi.

17

Bottazzi, G., G. Dosi, M. Lippi, F. Pammolli, and M. Riccaboni (2001). Innovation and corporate growth inthe evolution of the drug industry. International Journal of Industrial Organization 19 (7), 1161–1187.

Bottazzi, G. and A. Secchi (2006). Explaining the distribution of firm growth rates. Rand Journal of

Economics, 235–256.

Bottazzi, G. and A. Secchi (2011). A new class of asymmetric exponential power densities with applicationsto economics and finance. Industrial and Corporate Change, dtr036.

Buchinsky, M. (1998). Recent advances in quantile regression models: a practical guideline for empiricalresearch. Journal of human resources, 88–126.

Canay, I. A. (2011). A simple approach to quantile regression for panel data. The Econometrics Jour-

nal 14 (3), 368–386.

Casanova, L., F. Castellucci, J. Dayton-Johnson, S. Dutta, N. Fonstad, C. Paunov, and M. Pezzini (2011).Innovalatino: fostering innovation in latin america. insead.

Cassoni, A. and M. Ramada-Sarasola (2012). The returns to innovation in latin america: inexistent ormismeasured? Latin American Business Review 13 (2), 141–169.

Catela, E. Y., M. Cimoli, and G. Porcile (2015). Productivity and structural heterogeneity in the brazilianmanufacturing sector: trends and determinants. Oxford Development Studies (ahead-of-print), 1–21.

Chudnovsky, D., A. Lopez, and G. Pupato (2006). Innovation and productivity in developing countries: Astudy of argentine manufacturing firms behavior (1992–2001). Research policy 35 (2), 266–288.

Cimoli, M. et al. (2005). Heterogeneidad estructural, asimetrıas tecnologicas y crecimiento en america latina.

Cirillo, V. (2014). Patterns of innovation and wage distribution. do “innovative firms” pay higher wages?evidence from chile. Eurasian Business Review 4 (2), 181–206.

Coad, A. (2009). The growth of firms: A survey of theories and empirical evidence. Edward Elgar Publishing.

Coad, A., G. Pellegrino, and M. Savona (2015). Barriers to innovation and firm productivity. Economics of

Innovation and New Technology, 1–14.

Coad, A. and R. Rao (2008). Innovation and firm growth in high-tech sectors: A quantile regressionapproach. Research Policy 37 (4), 633–648.

Coad, A., A. Segarra, and M. Teruel (2013). Like milk or wine: Does firm performance improve with age?Structural Change and Economic Dynamics 24, 173–189.

Cohen, W. M. and S. Klepper (1996). A reprise of size and r&d. The Economic Journal , 925–951.

Colombelli, A., N. Haned, and C. Le Bas (2013). On firm growth and innovation: Some new empiricalperspectives using french cis (1992–2004). Structural Change and Economic Dynamics 26, 14–26.

Cozza, C., F. Malerba, M. L. Mancusi, G. Perani, and A. Vezzulli (2012). Innovation, profitability and growthin medium and high-tech manufacturing industries: evidence from italy. Applied Economics 44 (15),1963–1976.

Crepon, B., E. Duguet, and J. Mairessec (1998). Research, innovation and productivity: An econometricanalysis at the firm level. Economics of Innovation and new Technology 7 (2), 115–158.

Crespi, G. and P. Zuniga (2012). Innovation and productivity: evidence from six latin american countries.World Development 40 (2), 273–290.

Cucculelli, M. and B. Ermini (2012). New product introduction and product tenure: What effects on firmgrowth? Research Policy 41 (5), 808–821.

Damanpour, F., R. M. Walker, and C. N. Avellaneda (2009). Combinative effects of innovation typesand organizational performance: A longitudinal study of service organizations. Journal of Management

Studies 46 (4), 650–675.

18

Damijan, J. P., C. Kostevc, and M. Rojec (2012). Does innovation help the good or the poor performingfirms? Economics Letters 115 (2), 190–195.

Delmar, F., P. Davidsson, and W. B. Gartner (2003). Arriving at the high-growth firm. Journal of businessventuring 18 (2), 189–216.

Duguet, E. (2006). Innovation height, spillovers and tfp growth at the firm level: Evidence from frenchmanufacturing. Economics of Innovation and New Technology 15 (4-5), 415–442.

Ericson, R. and A. Pakes (1995). Markov-perfect industry dynamics: A framework for empirical work. The

Review of Economic Studies 62 (1), 53–82.

Eurostat (2013). Seventh community innovation survey. Available online at http://ec.europa.eu/

eurostat/web/science-technology-innovation/data/database.

Falk, M. (2012). Quantile estimates of the impact of r&d intensity on firm performance. Small Business

Economics 39 (1), 19–37.

Firpo, S. (2007). Efficient semiparametric estimation of quantile treatment effects. Econometrica, 259–276.

Fligner, M. A. and G. E. Policello (1981). Robust rank procedures for the behrens-fisher problem. Journal

of the American Statistical Association 76 (373), 162–168.

Freel, M. S. and P. J. Robson (2004). Small firm innovation, growth and performance evidence from scotlandand northern england. International Small Business Journal 22 (6), 561–575.

Frolich, M. and B. Melly (2010). Estimation of quantile treatment effects with stata. Stata Journal 10 (3),423.

Geroski, P. (2002). The growth of firms in theory and in prattice. Competence, governance and entrepreneur-

ship 1, 168–186.

Geroski, P. and S. Machin (1992). Do innovating firms outperform non-innovators?*. Business Strategy

Review 3 (2), 79–90.

Geroski, P. A. (1995). What do we know about entry? International Journal of Industrial Organiza-

tion 13 (4), 421–440.

Geroski, P. A., J. Van Reenen, and C. F. Walters (1997). How persistently do firms innovate? Research

Policy 26 (1), 33–48.

Gibrat, R. (1931). Les inegalites economiques. Recueil Sirey.

Goedhuys, M. and L. Sleuwaegen (2010). High-growth entrepreneurial firms in africa: a quantile regressionapproach. Small Business Economics 34 (1), 31–51.

Goedhuys, M. and L. Sleuwaegen (2016). High-growth versus declining firms: the differential impact ofhuman capital and r&d. Applied Economics Letters 23 (5), 369–372.

Goedhuys, M. and R. Veugelers (2012). Innovation strategies, process and product innovations and growth:Firm-level evidence from brazil. Structural Change and Economic Dynamics 23 (4), 516–529.

Hall, B. H. and J. Mairesse (2006). Empirical studies of innovation in the knowledge-driven economy.Economics of Innovation and New Technology 15 (4-5), 289–299.

Haltiwanger, J., R. S. Jarmin, and J. Miranda (2013). Who creates jobs? small versus large versus young.Review of Economics and Statistics 95 (2), 347–361.

Holzl, W. (2009). Is the r&d behaviour of fast-growing smes different? evidence from cis iii data for 16countries. Small Business Economics 33 (1), 59–75.

Holzl, W. and K. S. Friesenbichler (2010). High-growth firms, innovation and the distance to the frontier.Economics Bulletin 30 (2), 1016–1024.

19

Jovanovic, B. (1982). Selection and the evolution of industry. Econometrica: Journal of the Econometric

Society, 649–670.

Kannebley Jr, S., J. V. Sekkel, and B. C. Araujo (2010). Economic performance of brazilian manufacturingfirms: a counterfactual analysis of innovation impacts. Small Business Economics 34 (3), 339–353.

Koenker, R. and G. Bassett Jr (1978). Regression quantiles. Econometrica: journal of the Econometric

Society, 33–50.

Lever, M. and H. Nieuwenhuijsen (1999). The impact of competition on productivity in dutch manufacturing.Innovation, Industry Evolution and Employment , 111–128.

Mairesse, J. and P. Mohnen (2010). Using innovation surveys for econometric analysis. Handbook of the

Economics of Innovation 2, 1129–1155.

Mansfield, E. (1962). Entry, gibrat’s law, innovation, and the growth of firms. The American Economic

Review , 1023–1051.

Mazzucato, M. and S. Parris (2015). High-growth firms in changing competitive environments: the uspharmaceutical industry (1963 to 2002). Small Business Economics 44 (1), 145–170.

Ministerio de Economia (2014). Octava encuesta de innovacion en empresas, 2011-2012. princi-pales resultados. Available online at http://www.economia.gob.cl/wp-content/uploads/2014/02/

Presentacion-Resultados-8va-Encuesta-Innovacion1.pdf.

Moreno, F. and A. Coad (2015). High-growth firms: Stylized facts and conflicting results. Technical report,SPRU-Science and Technology Policy Research, University of Sussex.

Nelson, R. R. and S. Winter (1982). An evolutionary theory of economic change. An evolutionary theory of

economic change.

Nickell, S. J. (1996). Competition and corporate performance. Journal of political economy, 724–746.

OECD (2005). Oslo manual: Guidelines for collecting and interpreting innovation data. Number 4. Publi-cations de l’OCDE.

OECD/ECLAC (2013). Latin american economic outlook 2013.

OECD/World Bank (2015). The innovation policy platform. Available online at http://www.

innovationpolicyplatform.org/content/statistics-ipp.

Parente, P. M. and J. Santos Silva (2016). Quantile regression with clustered data. Journal of Econometric

Methods 5 (1), 1–15.

Paunov, C. (2012). The global crisis and firms investments in innovation. Research Policy 41 (1), 24–35.

Pavitt, K. (1984). Sectoral patterns of technical change: towards a taxonomy and a theory. Research

policy 13 (6), 343–373.

Perez, P., G. Dutrenit, and F. Barceinas (2005). Actividad innovadora y desempeno economico: un analisiseconometrico del caso mexicano. Indicadores de Ciencia y Tecnologıa en Iberoamerica. Buenos Aires,

Argentina: RICYT .

Pinto, A. (1970). Naturaleza e implicaciones de la” heterogeneidad estructural” de la america latina. El

trimestre economico, 83–100.

Raffo, J., S. Lhuillery, and L. Miotti (2008). Northern and southern innovativity: a comparison acrosseuropean and latin american countries. The European Journal of Development Research 20 (2), 219–239.

Santoleri, P. (2015). Diversity and intensity of information and communication technologies use and productinnovation: evidence from chilean micro-data. Economics of Innovation and New Technology 24 (6), 550–568.

Santoleri, P. and G. Stumpo (2011). Microempresas y pymes en amrica latina: Caractersticas de las firmasy polticas de apoyo. Documento de trabajo, Divisin de Desarrollo Productivo y Empresarial .

20

Sunkel, O. (1978). La dependencia y la heterogeneidad estructural. El trimestre economico, 3–20.

Vivarelli, M. (2014). Innovation, employment and skills in advanced and developing countries: a survey ofeconomic literature. Journal of Economic Issues 48 (1), 123–154.

8 Appendix

Table 7: Sample distribution by industry and size

Micro Small Medium Large Sample % (N)

Agriculture 1.25 1.90 5.32 1.86 2.55Mining 1.25 3.02 5.17 2.92 3.30Manufacturing 15.42 14.54 12.36 17.07 15.29Retail 25.66 19.35 13.65 25.17 20.29Transportation 8.33 10.96 10.63 9.03 9.83Real Estate 12.67 14.53 10.20 10.49 11.33Construction 10.83 11.19 8.19 11.09 10.49Electricity 0 0 3.74 5.51 3.27Hotel 16.67 15.88 12.07 4.91 10.19Finance 0 0 5.60 7.10 4.38Other services 7.92 7.94 7.47 4.85 6.44

Sample % (N) 7.19 (240) 26.80 (894) 20.86 (696) 45.14 (1,506) 100 (3336)

Source: Authors’ elaboration based on ELE.

Table 8: Summary Statistics and Variable Description

Variable Description Mean SD Min. Max. N

Sales gr Average annual firm growth (eq.1) 0.023 0.276 -3.145 1.181 3336Sales corr Average annual firm growth (eq.5) 0.032 0.263 -1.597 1.333 3336InnoProd =1 if firm introduced new or improved good or service (0/1) 0.274 0.446 0 1 3336InnoProc =1 if firm introduced a process innovation (0/1) 0.24 0.427 0 1 3336Foreign =1 if firm is at least partly owned by foreign capital 0.09 0.287 0 1 3336Expint Exports over total sales (%) 4.881 17.837 0 100 3336Group =1 if firm is a part of a business group (0/1) 0.277 0.448 0 1 3336Employment Number of employees (log) 3.669 2.1 0 11.124 3336Age Firm’s age (log) 2.593 0.712 0 5.043 3336Herfin Sum of the squared market share of each firm at the sector level 0.093 0.08 0.026 0.397 3336Public =1 if firm received public support (0/1) 0.182 0.386 0 1 3336HC Employees with tertiary education over total employees (%) 17.647 25.66 0 100 3336

21

Table 9: Pair-wise correlation matrixVariables Sales gr Sales corr InnoProdt−1 InnoProct−1 Foreignt−1 Expintt−1

Sales gr 1.000

Sales corr 0.989 1.000(0.000)

InnoProdt−1 -0.004 -0.007 1.000

(0.806) (0.667)InnoProct−1 0.030 0.028 0.377 1.000

(0.082) (0.104) (0.000)Foreignt−1 -0.006 -0.006 0.076 0.086 1.000

(0.724) (0.737) (0.000) (0.000)Expintt−1 -0.032 -0.031 0.056 0.065 0.164 1.000

(0.062) (0.069) (0.001) (0.000) (0.000)Groupt−1 0.019 0.017 0.141 0.185 0.282 0.137

(0.285) (0.339) (0.000) (0.000) (0.000) (0.000)Employmentt−1 0.010 0.001 0.215 0.286 0.238 0.179

(0.548) (0.953) (0.000) (0.000) (0.000) (0.000)Aget−1 -0.051 -0.061 0.096 0.101 0.067 0.033

(0.003) (0.000) (0.000) (0.000) (0.000) (0.056)HCt−1 -0.007 -0.005 0.068 0.045 0.211 0.056

(0.701) (0.768) (0.000) (0.010) (0.000) (0.001)Publict−1 0.024 0.020 0.120 0.186 0.082 0.116

(0.159) (0.250) (0.000) (0.000) (0.000) (0.000)Herfint−1 0.000 -0.004 0.107 0.065 0.040 0.150

(0.996) (0.806) (0.000) (0.000) (0.021) (0.000)

Variables Groupt−1 Employmentt−1 Aget−1 HCt−1 Publict−1 Herfint−1

Groupt−1 1.000

(0.000)Employmentt−1 0.412 1.000

(0.000) (0.000)Aget−1 0.109 0.284 1.000

(0.056) (0.000) (0.000)HCt−1 0.234 0.037 0.020 1.000

(0.001) (0.000) (0.034) (0.251)Publict−1 0.207 0.324 0.120 0.082 1.000

(0.000) (0.000) (0.000) (0.000) (0.000)Herfint−1 0.012 0.072 0.078 -0.104 0.077 1.000

(0.000) (0.471) (0.000) (0.000) (0.000) (0.000)

8.1 Robustness Check

In this section we present some additional estimates in order to check the robustness of theresults showed in Section 5.1. As Audretsch et al. (2014) pointed out, the indicators usedto measure growth are not neutral with respect to empirical results. Different measuresdescribe different patterns of growth (Delmar et al., 2003). For this reason, we introduce analternative growth rate measure to conduct a robustness check. Following Birch (1987) andSchreyer (2000), we use an index that helps reduce the bias toward larger firms (absolutegrowth) and small firms (relative growth rate). Proportional growth is biased towards smallfirms, as small units are much more likely to exhibit high rates of proportional growth thanlarge firms. Absolute growth is the absolute change in size. Measuring growth in absolute orrelative terms can lead to different results (Audretsch et al., 2014; Holzl and Friesenbichler,2010; Delmar et al., 2003). The index used is a simple combination of proportional andabsolute growth (Coad, 2009; Holzl and Friesenbichler, 2010):

Gri,t =(ln(Si,t)− ln(Si,t−d)) ·

ln(Si,t)ln(Si,t−d)

d(5)

This measure is still dependent on firm size, but has a smaller bias toward firm size ifcompared with proportional or absolute measures of growth.

As a first step, we also check the Fligner-Policello test using this alternative growth ratemeasure and results do not change.

Successively, we estimate the models presented in Section 5.1 employing (5) as our de-pendent variable. Results are shown in Tables 11, 12 and 13. We can observe that, with

22

Table 10: Two-sample Fligner-Policello robust rank order test

Variable Obs Mean Index of Variability U 2-Tailed p-value

InnoProdt−1 No 2422 4.5e+02 1.9e+08 -0.383 0.70180InnoProdt−1 Yes 914 1.2e+03 4.0e+08

InnoProct−1 No 2537 3.8e+02 1.4e+08 -2.471 0.01348InnoProct−1 Yes 799 1.3e+03 4.3e+08

Source: Authors’ elaboration based on ELE. Growth is defined as in (5).

respect to the innovation variables, we do not appreciate any significant change. Indeed,point estimates and patterns of significance show little variation compared to the estimatesobtained using (5).

Table 11: OLS, FE and QR - Innovation and Sales Growth(1) (2) (3) (4) (5) (6) (7)OLS FE q10 q25 q50 q75 q90

InnoProdt−1 -0.013 -0.039** -0.033 0.002 -0.005 -0.002 -0.023(0.011) (0.017) (0.022) (0.011) (0.007) (0.009) (0.015)

InnoProct−1 0.026** 0.015 0.047** 0.017 0.015* 0.027** 0.045***(0.012) (0.017) (0.019) (0.012) (0.008) (0.010) (0.017)

Employmentt−1 0.000 -0.065*** 0.019*** 0.012*** 0.005*** -0.003 -0.012***(0.003) (0.013) (0.005) (0.003) (0.002) (0.002) (0.004)

Aget−1 -0.026*** -0.020 0.017 0.010 -0.010** -0.039*** -0.068***(0.008) (0.022) (0.016) (0.008) (0.005) (0.006) (0.012)

Foreignt−1 -0.001 0.008 0.001 -0.010 0.006 0.031** 0.057**(0.018) (0.038) (0.034) (0.020) (0.013) (0.015) (0.027)

Expintt−1 -0.000 -0.000 -0.001*** -0.001** -0.000** -0.000 0.001(0.000) (0.001) (0.001) (0.000) (0.000) (0.000) (0.001)

Groupt−1 0.009 -0.047 -0.032 -0.022* -0.009 0.010 0.025(0.012) (0.031) (0.024) (0.012) (0.008) (0.010) (0.017)

HCt−1 -0.000 -0.001 -0.000 0.000 0.000 0.000 -0.000(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Publict−1 0.016 0.014 0.035* 0.016 -0.003 0.002 -0.004(0.011) (0.020) (0.019) (0.011) (0.007) (0.010) (0.018)

Herfint−1 -0.030 -0.132 -0.102 -0.108 0.038 0.010 -0.724(0.276) (0.292) (0.591) (0.337) (0.180) (0.224) (0.485)

Constant 0.057 0.359*** -0.438*** -0.233*** -0.014 0.205*** 0.638***(0.050) (0.112) (0.099) (0.049) (0.033) (0.045) (0.088)

Observations 3,336 3,336 3,336 3,336 3,336 3,336 3,336R2 0.010 0.042 0.046 0.023 0.006 0.022 0.052

Notes: Robust standard errors and boostrapped estimates (1500 iterations). ***, ** and * indicatesignificance on a 1%, 5% and 10% level, respectively. Industry and year dummies are included but not

reported.

23

Table 12: Average Matching EstimatesAverage Treatment Effects on the Treated - PSM

Coefficient AI SE z P > |z|

Product Innovators -0.00321 0.0143 -0.22 0.823Process Innovators 0.02432* 0.0132 1.83 0.067

Average Treatment Effects on the Treated - NNMCoefficient AI SE z P > |z|

Product Innovators -0.02614* 0.0143 -1.83 0.068Process Innovators 0.03710** 0.0153 2.41 0.016

Average Treatment Effects - PSMCoefficient AI SE z P > |z|

Product Innovators -0.00815 0.0142 -0.57 0.566Process Innovators 0.03091* 0.0181 1.70 0.089

Average Treatment Effects - NNMCoefficient AI SE z P > |z|

Product Innovators -0.02068 0.0139 -1.48 0.138Process Innovators 0.05193*** 0.0154 3.37 0.001

Notes: ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively.

Table 13: Unconditional Quantile Treatment Effect Estimatesq10 q25 q50 q75 q90

Product Innovators -0.0082 0.0073 -0.0043 -0.0138 -0.0381*(0.0300) (0.0136) (0.0085) (0.0115) (0.0191)

Process Innovators 0.0144 0.0210 0.0184 0.0502** 0.0459*(0.0257) (0.0189) (0.0121) (0.0172) (0.0257)

Notes: Standard errors in parenthesis. ***, ** and * indicate significance on a 1%, 5% and 10% level,respectively.

24